【摘要】 在基于语音识别的计算机辅助发音学习(CAPT)系统rEcho V1.0的基础上设计并实现了rEcho V2.0.系统继承了V1.0利用错误趋势分组技术为学习者提供有针对性反馈建议的优点,完善了课程体系,提出利用共振峰错音合成技术合成识别器训练所需错音语料的方法,为解决错音语料不足这一长期困扰CAPT的问题提供了有效的途径.实验表明,合成错音训练的识别器达到原始语音训练的识别器的识别率水平.rEcho V2.0的整体性能较之前版本有较大提高.更多还原

【Abstract】 Speech recognition-based computer assisted pronunciation training (CAPT) system is a promising technology. CAPT system rEcho 2.0 was developed from an early version of rEcho V1.0. rEcho 2.0 outperforms rEcho V1.0 in that it could give suggestions to improve learner’s pronunciation (by error trend grouping),deliver better learning content and schedule,train error phone acoustic models by synthesis of error speech. The method of error sound formant synthesis gives a way to solve the problem of error model training data shortage. Experiment showed that the synthesis speech can give satisfied recognition results as the pre-recorded speech does.更多还原